diff --git a/CMakeLists.txt b/CMakeLists.txt
index b7288df348781310118754a1598614884950fae6..a35babdceb29a16e5a5f3dc1cd251e6a959b5dce 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -172,31 +172,31 @@ endif()
# Use intermediate variable since cmake_dependent_option doesn't have OR conditions
set(CHAMELEON_ENABLE_MPI OFF CACHE INTERNAL "Tells if MPI might be supported by the runtime")
-if ( CHAMELEON_SCHED_PARSEC OR CHAMELEON_SCHED_STARPU )
- set(CHAMELEON_ENABLE_MPI ON FORCE)
+if ( CHAMELEON_SCHED_STARPU )
+ set(CHAMELEON_ENABLE_MPI ON FORCE)
endif()
# Use intermediate variable since cmake_dependent_option doesn't have OR conditions
set(CHAMELEON_ENABLE_CUDA OFF CACHE INTERNAL "Tells if CUDA might be supported by the runtime")
if ( CHAMELEON_SCHED_PARSEC OR CHAMELEON_SCHED_STARPU )
- set(CHAMELEON_ENABLE_CUDA ON FORCE)
+ set(CHAMELEON_ENABLE_CUDA ON FORCE)
endif()
set(CHAMELEON_ENABLE_HIP OFF CACHE INTERNAL "Tells if HIP might be supported by the runtime")
if ( CHAMELEON_SCHED_STARPU )
- set(CHAMELEON_ENABLE_HIP ON FORCE)
+ set(CHAMELEON_ENABLE_HIP ON FORCE)
endif()
# Additional options
# ------------------
-# Enable the distributed interface (allowed only when StarPU or PaRSEC is enabled)
+# Enable the distributed interface (allowed only when StarPU is enabled)
# TODO: Default should be changed to ON/OFF when it will be ok
cmake_dependent_option(CHAMELEON_USE_MPI
"Enable distributed memory through MPI" OFF
"CHAMELEON_ENABLE_MPI" OFF)
if (CHAMELEON_ENABLE_MPI AND NOT CHAMELEON_USE_MPI)
- message("-- ${BoldGreen}CHAMELEON_USE_MPI is set to OFF, turn it ON to use MPI (unsupported by Quark)${ColourReset}")
+ message("-- ${BoldGreen}CHAMELEON_USE_MPI is set to OFF, turn it ON to use MPI (only supported with StarPU)${ColourReset}")
endif()
cmake_dependent_option(CHAMELEON_USE_MIGRATE
@@ -213,7 +213,7 @@ cmake_dependent_option(CHAMELEON_USE_CUDA
"Enable CUDA kernels" OFF
"CHAMELEON_ENABLE_CUDA" OFF)
if (CHAMELEON_ENABLE_CUDA AND NOT CHAMELEON_USE_CUDA)
- message("-- ${BoldGreen}CHAMELEON_USE_CUDA is set to OFF, turn it ON to use CUDA (unsupported by Quark)${ColourReset}")
+ message("-- ${BoldGreen}CHAMELEON_USE_CUDA is set to OFF, turn it ON to use CUDA (unsupported by OpenMP and Quark)${ColourReset}")
endif()
cmake_dependent_option(CHAMELEON_USE_HIP_CUDA
@@ -551,6 +551,11 @@ if(NOT CHAMELEON_SIMULATION)
morse_export_imported_target(MPI MPI_C mpi chameleon)
+ if (CHAMELEON_SCHED STREQUAL "PARSEC")
+ message(FATAL_ERROR "CHAMELEON does not support PARSEC with MPI."
+ " To use PARSEC set CHAMELEON_USE_MPI=OFF.")
+ endif()
+
endif (CHAMELEON_USE_MPI)
else (NOT CHAMELEON_SIMULATION)
diff --git a/doc/user/chameleon.dot b/doc/user/chameleon.dot
index 6cbc54d892d922b74c23952fe953084a9fe6f03c..99b65026d47ca04a583865193092336d72abc411 100644
--- a/doc/user/chameleon.dot
+++ b/doc/user/chameleon.dot
@@ -27,7 +27,6 @@ chameleon -> sched [lhead=cluster_runtime]
sched:sched_starpu -> cuda [color=dimgrey]
sched:sched_starpu -> mpi [color=dimgrey]
sched:sched_parsec -> cuda [color=dimgrey]
-sched:sched_parsec -> mpi [color=dimgrey]
subgraph cluster_paradigm {
label = "Paradigms"
diff --git a/doc/user/chameleon.svg b/doc/user/chameleon.svg
index 4a51dd62fcb01bd34df3e301af8ce6c7c6d61016..d1f9c6fc5d16720a383e7bc34f3fe10a5731e7bd 100644
--- a/doc/user/chameleon.svg
+++ b/doc/user/chameleon.svg
@@ -15,8 +15,8 @@
</g>
<g id="clust2" class="cluster">
<title>cluster_paradigm</title>
-<path fill="transparent" stroke="black" d="M191,-8C191,-8 317,-8 317,-8 323,-8 329,-14 329,-20 329,-20 329,-71 329,-71 329,-77 323,-83 317,-83 317,-83 191,-83 191,-83 185,-83 179,-77 179,-71 179,-71 179,-20 179,-20 179,-14 185,-8 191,-8"/>
-<text text-anchor="middle" x="254" y="-67.8" font-family="Times,serif" font-size="14.00">Paradigms</text>
+<path fill="transparent" stroke="black" d="M208,-8C208,-8 334,-8 334,-8 340,-8 346,-14 346,-20 346,-20 346,-71 346,-71 346,-77 340,-83 334,-83 334,-83 208,-83 208,-83 202,-83 196,-77 196,-71 196,-71 196,-20 196,-20 196,-14 202,-8 208,-8"/>
+<text text-anchor="middle" x="271" y="-67.8" font-family="Times,serif" font-size="14.00">Paradigms</text>
</g>
<g id="clust3" class="cluster">
<title>cluster_kernel</title>
@@ -50,14 +50,14 @@
<!-- mpi -->
<g id="node4" class="node">
<title>mpi</title>
-<polygon fill="none" stroke="gold" stroke-width="2" points="241,-52 187,-52 187,-16 241,-16 241,-52"/>
-<text text-anchor="start" x="197" y="-31.3" font-family="Times,serif" font-weight="bold" font-size="14.00">MPI</text>
+<polygon fill="none" stroke="gold" stroke-width="2" points="338,-52 284,-52 284,-16 338,-16 338,-52"/>
+<text text-anchor="start" x="294" y="-31.3" font-family="Times,serif" font-weight="bold" font-size="14.00">MPI</text>
</g>
<!-- chameleon->mpi -->
-<g id="edge10" class="edge">
+<g id="edge9" class="edge">
<title>chameleon->mpi</title>
-<path fill="none" stroke="dimgrey" d="M359.38,-174.94C355.97,-172.58 352.78,-169.94 350,-167 324.84,-140.35 347.29,-113.02 318,-91 296.72,-75 280.81,-92.5 259.5,-86.95"/>
-<polygon fill="dimgrey" stroke="dimgrey" points="260.58,-83.61 250,-83 257.89,-90.07 260.58,-83.61"/>
+<path fill="none" stroke="dimgrey" d="M358.3,-174.72C355.24,-172.43 352.42,-169.86 350,-167 340.22,-155.41 330.55,-122.87 323.28,-92.99"/>
+<polygon fill="dimgrey" stroke="dimgrey" points="326.63,-91.92 320.92,-83 319.82,-93.54 326.63,-91.92"/>
</g>
<!-- cublas -->
<g id="node5" class="node">
@@ -66,7 +66,7 @@
<text text-anchor="start" x="370.5" y="-114.8" font-family="Times,serif" font-weight="bold" font-size="14.00">cuBLAS</text>
</g>
<!-- chameleon->cublas -->
-<g id="edge11" class="edge">
+<g id="edge10" class="edge">
<title>chameleon->cublas</title>
<path fill="none" stroke="black" d="M401,-174.95C401,-166.3 401,-155.57 401,-145.79"/>
<polygon fill="black" stroke="black" points="404.5,-145.71 401,-135.71 397.5,-145.71 404.5,-145.71"/>
@@ -78,7 +78,7 @@
<text text-anchor="start" x="465.5" y="-114.8" font-family="Times,serif" font-weight="bold" font-size="14.00">LAPACKE</text>
</g>
<!-- chameleon->lapacke -->
-<g id="edge9" class="edge">
+<g id="edge8" class="edge">
<title>chameleon->lapacke</title>
<path fill="none" stroke="black" d="M436.21,-174.98C440.62,-172.47 444.98,-169.78 449,-167 459.25,-159.89 469.65,-151.03 478.55,-142.83"/>
<polygon fill="black" stroke="black" points="481.07,-145.27 485.94,-135.86 476.27,-140.18 481.07,-145.27"/>
@@ -90,7 +90,7 @@
<text text-anchor="start" x="481" y="-31.3" font-family="Times,serif" font-weight="bold" font-size="14.00">CBLAS</text>
</g>
<!-- chameleon->cblas -->
-<g id="edge8" class="edge">
+<g id="edge7" class="edge">
<title>chameleon->cblas</title>
<path fill="none" stroke="black" d="M453.77,-190.72C494,-188.5 544.7,-182.59 558,-167 579.93,-141.31 570.04,-122.56 558,-91 553.46,-79.11 545.11,-68.14 536.52,-59.12"/>
<polygon fill="black" stroke="black" points="538.9,-56.55 529.32,-52.03 533.98,-61.54 538.9,-56.55"/>
@@ -98,41 +98,35 @@
<!-- cuda -->
<g id="node3" class="node">
<title>cuda</title>
-<polygon fill="none" stroke="gold" stroke-width="2" points="321,-52 259,-52 259,-16 321,-16 321,-52"/>
-<text text-anchor="start" x="267" y="-31.3" font-family="Times,serif" font-weight="bold" font-size="14.00">CUDA</text>
+<polygon fill="none" stroke="gold" stroke-width="2" points="266,-52 204,-52 204,-16 266,-16 266,-52"/>
+<text text-anchor="start" x="212" y="-31.3" font-family="Times,serif" font-weight="bold" font-size="14.00">CUDA</text>
</g>
<!-- sched->cuda -->
<g id="edge2" class="edge">
<title>sched:sched_starpu->cuda</title>
-<path fill="none" stroke="dimgrey" d="M273,-98.5C273,-86.18 276.02,-72.96 279.52,-61.79"/>
-<polygon fill="dimgrey" stroke="dimgrey" points="282.88,-62.78 282.79,-52.19 276.25,-60.52 282.88,-62.78"/>
+<path fill="none" stroke="dimgrey" d="M273,-98.5C273,-84.71 266.32,-71.33 258.57,-60.43"/>
+<polygon fill="dimgrey" stroke="dimgrey" points="261.12,-58.01 252.22,-52.26 255.59,-62.3 261.12,-58.01"/>
</g>
<!-- sched->cuda -->
<g id="edge4" class="edge">
<title>sched:sched_parsec->cuda</title>
-<path fill="none" stroke="dimgrey" d="M131,-98.5C131,-71.83 226.63,-95.85 250,-83 259.7,-77.67 268.01,-68.97 274.52,-60.42"/>
-<polygon fill="dimgrey" stroke="dimgrey" points="277.5,-62.26 280.39,-52.07 271.78,-58.23 277.5,-62.26"/>
+<path fill="none" stroke="dimgrey" d="M131,-98.5C131,-67.65 164.74,-51.35 193.78,-43.02"/>
+<polygon fill="dimgrey" stroke="dimgrey" points="195.01,-46.32 203.8,-40.4 193.24,-39.55 195.01,-46.32"/>
</g>
<!-- sched->mpi -->
<g id="edge3" class="edge">
<title>sched:sched_starpu->mpi</title>
-<path fill="none" stroke="dimgrey" d="M273,-98.5C273,-86.17 258.93,-91.49 250,-83 243.05,-76.4 236.45,-68.34 230.84,-60.73"/>
-<polygon fill="dimgrey" stroke="dimgrey" points="233.38,-58.26 224.75,-52.11 227.66,-62.3 233.38,-58.26"/>
-</g>
-<!-- sched->mpi -->
-<g id="edge5" class="edge">
-<title>sched:sched_parsec->mpi</title>
-<path fill="none" stroke="dimgrey" d="M131,-98.5C131,-73.42 155.21,-56.91 177.47,-47.02"/>
-<polygon fill="dimgrey" stroke="dimgrey" points="178.9,-50.22 186.82,-43.19 176.24,-43.75 178.9,-50.22"/>
+<path fill="none" stroke="dimgrey" d="M273,-98.5C273,-84.71 279.68,-71.33 287.43,-60.43"/>
+<polygon fill="dimgrey" stroke="dimgrey" points="290.41,-62.3 293.78,-52.26 284.88,-58.01 290.41,-62.3"/>
</g>
<!-- cublas->cuda -->
-<g id="edge7" class="edge">
+<g id="edge6" class="edge">
<title>cublas->cuda</title>
-<path fill="none" stroke="black" d="M373.95,-99.4C366.12,-94.28 357.63,-88.54 350,-83 339.46,-75.35 328.23,-66.55 318.37,-58.59"/>
-<polygon fill="black" stroke="black" points="320.42,-55.75 310.45,-52.15 316,-61.18 320.42,-55.75"/>
+<path fill="none" stroke="black" d="M362.41,-103.64C348.51,-99.28 332.64,-94.6 318,-91 299.12,-86.36 291.77,-92.83 275,-83 265.59,-77.49 257.41,-68.91 250.93,-60.52"/>
+<polygon fill="black" stroke="black" points="253.72,-58.39 245.05,-52.31 248.03,-62.47 253.72,-58.39"/>
</g>
<!-- lapacke->cblas -->
-<g id="edge6" class="edge">
+<g id="edge5" class="edge">
<title>lapacke->cblas</title>
<path fill="none" stroke="black" d="M504.06,-99.22C504.72,-88.52 505.57,-74.55 506.32,-62.37"/>
<polygon fill="black" stroke="black" points="509.82,-62.45 506.94,-52.26 502.84,-62.03 509.82,-62.45"/>
diff --git a/doc/user/chapters/installing.org b/doc/user/chapters/installing.org
index c28bee958a769afd6e0d4f857a53f1bb9b756f22..de4e616663832344647386f10e9ac4b481bc7e1f 100644
--- a/doc/user/chapters/installing.org
+++ b/doc/user/chapters/installing.org
@@ -208,7 +208,7 @@ Finally some packages or also available for [[sec:ug:debian][Debian/Ubuntu]] and
***** PaRSEC
[[http://icl.utk.edu/parsec/][PaRSEC]] is a generic framework for architecture aware scheduling
and management of micro-tasks on distributed many-core
- heterogeneous architectures. It can be used with MPI and Cuda.
+ heterogeneous architectures.
*Caution about the compatibility:* Chameleon is compatible with
this version
@@ -217,9 +217,9 @@ Finally some packages or also available for [[sec:ug:debian][Debian/Ubuntu]] and
[[http://icl.cs.utk.edu/quark/][QUARK]] (QUeuing And Runtime for Kernels) provides a library that
enables the dynamic execution of tasks with data dependencies in
a multi-core, multi-socket, shared-memory environment. When
- Chameleon is linked with QUARK, it is not possible to exploit
+ Chameleon is linked with QUARK or OPENMP, it is not possible to exploit
neither CUDA (for GPUs) nor MPI (distributed-memory environment).
- You can use PaRSEC or StarPU to do so.
+ You can use StarPU to do so.
*Caution about the compatibility:* Chameleon has been mainly tested
with the QUARK library coming from https://github.com/ecrc/quark.
@@ -366,7 +366,6 @@ Finally some packages or also available for [[sec:ug:debian][Debian/Ubuntu]] and
* *CHAMELEON_USE_MPI=ON|OFF* (default OFF): to link with MPI
library (message passing implementation for use of multiple
nodes with distributed memory), can only be used with StarPU
- and PaRSEC
* *CHAMELEON_USE_CUDA=ON|OFF* (default OFF): to link with CUDA
runtime (implementation paradigm for accelerated codes on Nvidia GPUs)
and cuBLAS library (optimized BLAS kernels on Nvidia GPUs), can only
@@ -460,6 +459,9 @@ Finally some packages or also available for [[sec:ug:debian][Debian/Ubuntu]] and
Note that PaRSEC and StarPU are only detected with pkg-config
mechanism because it is always provided and this avoids errors.
+ The [[https://cmake.org/cmake/help/latest/variable/CMAKE_PREFIX_PATH.html#variable:CMAKE_PREFIX_PATH][CMAKE_PREFIX_PATH]]
+ can be used to indicate where dependencies are installed.
+
*** Distribution Debian
:PROPERTIES:
:CUSTOM_ID: doc-install-debian
diff --git a/doc/user/chapters/introduction.org b/doc/user/chapters/introduction.org
index 86821edecebffef514ab0b73bab7d8bc6b13cec4..8802d35c3fcc67219fde0666dbca5b567b2a81ed 100644
--- a/doc/user/chapters/introduction.org
+++ b/doc/user/chapters/introduction.org
@@ -139,11 +139,14 @@
Chameleon is based on the [[http://icl.cs.utk.edu/plasma/][PLASMA]] source code but is not limited to
shared-memory environment and can exploit multiple GPUs.
- Chameleon is interfaced in a generic way with [[http://runtime.bordeaux.inria.fr/StarPU/][StarPU]], [[http://icl.utk.edu/parsec/][PaRSEC]],
- [[http://icl.cs.utk.edu/quark/][QUARK]] runtime systems. This feature allows to analyze in a
- unified framework how sequential task-based algorithms behave
+ Chameleon is interfaced in a generic way with
+ [[http://runtime.bordeaux.inria.fr/StarPU/][StarPU]],
+ [[http://icl.utk.edu/parsec/][PaRSEC]],
+ [[https://www.openmp.org/][OpenMP]],
+ [[http://icl.cs.utk.edu/quark/][QUARK]] runtime systems.
+ This feature allows to analyze in a unified framework how sequential task-based algorithms behave
regarding different runtime systems implementations. Using
- Chameleon with *StarPU* or *PaRSEC* runtime systems allows to exploit
+ Chameleon with *StarPU* runtime system allows to exploit
GPUs through kernels provided by [[https://developer.nvidia.com/cublas][cuBLAS]] and clusters of
interconnected nodes with distributed memory (using [[http://www.open-mpi.org/][MPI]]).
Computation of very large systems with dense matrices on a cluster
diff --git a/doc/user/homepage.org b/doc/user/homepage.org
index f2a364273bce421c2f83f3dc7db968c10142d5cb..b00bde6dc1f35379bcf09b5efa6419279969347c 100644
--- a/doc/user/homepage.org
+++ b/doc/user/homepage.org
@@ -27,7 +27,7 @@
StarPU, PaRSEC, QUARK, OpenMP runtime systems. This feature allows
to analyze in a unified framework how sequential task-based
algorithms behave regarding different runtime systems
- implementations. Using Chameleon with StarPU or PaRSEC runtime
+ implementations. Using Chameleon with StarPU runtime
systems allows to exploit GPUs through kernels provided by cuBLAS
and clusters of interconnected nodes with distributed memory (using
MPI).